1908: EL CENTENARIO DE LOS SITIOS DE ZARAGOZA Y LA EXPOSICIÓN HISPANOFRANCESA.

Introduction

Much research on the early detection of cognitive impairment and dementia has concentrated on the assessment of memory and other cognitive domain performance. However, there is growing interest in the use of other indicators of cognitive function, such as processing speed measures. A meta-analysis conducted to determine the characteristics of cognitive domain impairment in preclinical Alzheimer’s disease demonstrated that, in addition to the well-known deficits in episodic memory, losses in processing speed also occur early in the disease process (Backman, Jones, Berger, Laukka, & Small, 2005). A simple way of assessing processing speed is to measure the time it takes to react to a stimulus and this Chapter presents work which was conducted to evaluate the effectiveness of reaction time task derived measures in identifying MCI. These measures were selected as further potentially useful, simple tests that could provide an alternative to the traditionally used memory tests described previously. Measures based on reaction time tasks have an added benefit in that they are unlikely to be influenced by administrator bias, education level and language ability like many of the aforementioned tests.

Previous research has demonstrated that mean reaction times tend to increase in normal ageing (Simon, 1968) and in pathological ageing, with studies showing increased reaction times in patients with cognitive impairment compared to unimpaired controls (Anstey et al., 2007; Dixon et al., 2007). Processing speed is thought to reflect

underlying neural integrity and it has been theorised that cognitive performance is degraded when processing speed is slow due to a limited availability of neural resources

to support higher level cognitive behaviour such as episodic memory (Salthouse, 1996). In their study, Anstey et al (2007) provided evidence to suggest that processing speed performance is related to brain structure, with their results demonstrating that faster reactions times were associated with larger corpus callosum size in both healthy controls and those with mild cognitive disorders.

As well as mean level of processing speed, the consistency with which an individual performs across trials within a task has also been suggested as being an important indicator of cognitive functioning (Hultsch & MacDonald, 2004; Jensen, 1992). The term most commonly used to describe a person’s level of consistency on a task is intra- individual variability (IIV) and it has been proposed that this measure may provide another behavioural marker of neurobiological disturbance (Hultsch, Strauss, Hunter, & MacDonald, 2008; MacDonald, Nyberg, & Backman, 2006). Previous research has demonstrated that IIV increases in normal ageing (for review see (Dykiert, Der, Starr, & Deary, 2012), traumatic brain injury (Hetherington, Stuss, & Finlayson, 1996; Stuss, Pogue, Buckle, & Bondar, 1994), mild cognitive impairment (Christensen et al., 2005; Dixon et al., 2007; Gorus, De Raedt, Lambert, Lemper, & Mets, 2008) and mild

dementia (Burton, Strauss, Hultsch, Moll, & Hunter, 2006; Hultsch et al., 2000). In their study, Hultsch et al (2000) demonstrated that IIV was greater in patients with mild dementia than in cognitively healthy elderly people, regardless of whether or not they had arthritis, indicating that increased IIV is probably primarily due to central

neurological rather than somatic disturbances. In addition, Burton et al (2006) demonstrated that IIV is most likely associated with specific, rather than general nervous system disturbances, with their finding that patients with AD were more inconsistent than those with Parkinson’s disease. Studies have also shown that IIV in reaction time predicts longitudinal cognitive decline in ageing populations (Bielak,

Hultsch, Strauss, Macdonald, & Hunter, 2010; Lovden, Li, Shing, & Lindenberger, 2007; MacDonald, Hultsch, & Dixon, 2003).

Increases in IIV in reaction time performance have been proposed to reflect the

increased neural noise and reduction in cortical representation that could result from the white matter decline that occurs in the brain during normal and pathological aging (MacDonald, Li, & Backman, 2009). Evidence to support this theory has been

demonstrated in MRI studies; for instance, Bunce et al. (2007) found that white matter lesioning, particularly in the frontal lobe, was associated with elevated IIV on simple reaction time tasks. In addition, Jackson et al (2012) found strong associations between IIV measures and total cerebral white matter volume, as well as frontal and parietal region volumes, in healthy older adults and participants with early stage Alzheimer’s disease. Increased IIV has been linked to frontal cortex mediated processes such as attentional lapses (Bunce, Warr, & Cochrane, 1993) and fluctuations in executive control (West, Murphy, Armilio, Craik, & Stuss, 2002).

Previous evidence suggests a link between cognition and performance on simple reaction time (RT) tasks, leading to the proposal that measures derived from RT tasks have the potential to aid the identification of a range of neurobiological disorders, including mild cognitive impairment (Bunce et al., 2013). This chapter reports the findings from a study that was conducted to investigate the use of RT task derived measures in identifying cognitive impairment.

Studies to date of RT performance in cognitive impairment and dementia have tended to use clinical samples in which the disease is likely to be relatively far progressed (Burton et al., 2006; Hultsch et al., 2000). Alternatively, studies investigating pre-dementia

phases have tended to use a rather broad definition of mild cognitive impairment that includes individuals with “age-associated memory impairment”, “aging-associated cognitive decline” and “mild neurocognitive disorder” (Anstey et al., 2007; Christensen et al., 2005) or have focussed on only the amnestic form of MCI (Gorus et al., 2008). This study aimed to evaluate the use of RT task derived measures within a well- characterised sample of community-dwelling participants including healthy controls, people with amnestic and non-amnestic MCI and people with cognitive difficulties beyond MCI (i.e. possible early dementia). People with low mood were also included since it has been proposed that depression may affect cognitive performance and has been associated with increased IIV in RT performance (Bunce, Handley, & Gaines, 2008).

Participants were administered two RT tasks: (i) a simple, two choice RT task (2CRT), with button box response and (ii) a more complex five choice RT task (5CRT), with touchscreen response. The inclusion of RT tasks with varying complexity enables the influence of task complexity to be investigated. There is evidence to suggest that increasing task complexity is associated with poorer performance in cognitively

impaired groups (Dixon et al., 2007; Gorus et al., 2008; Hultsch et al., 2000); however, questions still remain as to the optimal level of complexity of RT task that should be used to assess RT performance in the identification of cognitive impairment (Bielak et al., 2010). Also, the use of touchscreen technology enabled spatial accuracy, as well as speed, of response to be measured, and thus provided an additional novel measure, currently unexplored (to the author’s knowledge) within this field of research. In summary, the current study aimed to investigate the effectiveness of a range of measures derived from RT tasks for identifying people with cognitive impairment, by addressing the following questions: (1) Are there differences in (a) mean RT and RT

variability and (b) accuracy measures between people depending on their cognitive classification?; (2) Does the complexity of the RT task (2CRT vs. 5CRT) have a

differential effect on RT performance that is dependent on cognitive classification?; and (3) Are RT task derived measures able to predict cognitive classification?

Methods Participants

Older people (aged 70 years and above) were invited to participate via flyers sent to them from their GP practice. They were part of a larger study cohort recruited to assess the validity of two brief cognitive tests (described previously in Chapter 4). All

participants were evaluated using the battery of cognitive assessments described in Chapter 3 and were categorised into the following groups: (1) Control; (2) Amnestic MCI (aMCI); (3) Non-amnestic MCI (naMCI); (4) Cognitive difficulties beyond MCI (>MCI) and (5) Low mood. A total of 225 people were included in this study. However, one person could not be categorised due to a hearing impairment that impacted on their performance in the cognitive assessments, and was therefore excluded, leaving a total of 224 people in the study.

Participants with cognitive impairment

Thirty-nine people met the Petersen criteria for MCI (Petersen, 2004). Of these, 28 demonstrated impairment in memory (defined as CVLT Short Delay and Long Delay free recall ≥ 1.5 standard deviations below mean of published norms (Delis et al., 2000)) and were classified as having aMCI (amnestic MCI). The remaining 11 people demonstrated impairment in other non-memory domain(s) and were classified as having naMCI (non-amnestic MCI). Both single domain and multi-domain MCI participants were included.

Nine people demonstrated cognitive impairment that had detrimental effects on activities of daily living (as measured by the informant-administered Bristol Activities of Daily Living Scale (Bucks et al., 1996)) and were therefore classified as having cognitive impairment beyond MCI (>MCI).

Participants with low mood

Twelve people scored ≥6 on the Geriatric Depression Scale (Sheikh & Yesavage, 1986) and were classified as having a low mood.

Control participants

The remaining 164 people did not meet any of the previously described classifications and were therefore classified as controls and formed the reference group for the subsequent analyses.

The study was approved by the Yorkshire and The Humber National Research Ethics Service Committee (ref: 12/YH/0207) and all participants gave informed written consent. Ethical approval was also granted from the University of Leeds for the administration of the RT tasks, which were developed and administered using University-owned equipment (ref: 13.0256).

Reaction Time Tests

Participants performed two reaction time (RT) tasks. One was a simple two choice RT task and the other was a more complex five choice RT task. Both tasks were designed in ePrime ® version 2.0.8.90 (Psychology Software Tools Inc., USA) and run on a

arm’s length (30-50cm) away from the screen when performing the tasks. The order in which the tasks were administered was alternated between participants.

Two Choice Reaction Time Task (2CRT)

This task involved participants responding to a black circular target randomly presented on either the left or right hand side of a white screen (see Figure 5.1a) by pressing the corresponding button on a button box. A CEDRUS ® RB-540 button box attached via USB port to the laptop was used in this task. The circular targets were 25mm in diameter and were presented with an inter-stimulus interval of 500ms, during which a black fixation cross appeared in the centre of the screen. The circular target was

positioned 4cm from the fixation cross and stayed on screen until the participant pressed a button. Participants were instructed to respond as quickly and as accurately as

possible. Participants completed 12 practice trials, followed by 48 experimental trials. Based on an imposed maximum time limit per trial of 10 seconds, the task did not take longer than 10 minutes to administer, with the majority of participants able to complete the task in under 5 minutes.

Five Choice Reaction Time Task (5CRT)

This task utilised the touchscreen function of the laptop. In this task, participants responded to black circular targets appearing on a white screen by touching the target. The circular targets were 20mm in diameter and were randomly presented around the screen in one of five possible positions, equidistant from the centre at 9cm (see Figure 5.1b). The inter-stimulus interval was 500 or 1000ms (determined randomly) and during this time a box containing the words “touch here” appeared at the bottom of the screen. Participants were instructed to begin each trial by pressing and holding the “touch here” box, ensuring that they began each trial from the same starting point. The circular target

stayed on screen until the participant touched the screen. Again, participants were instructed to respond as quickly and as accurately as possible. Participants completed 5 practice trials, followed by 50 experimental trials. Based on an imposed maximum time limit per trial of 10 seconds, the task did not take longer than 9 minutes and 10 seconds to administer, with the majority of participants able to complete the task in under 5 minutes.

Figure 5.1: Possible target positions on the (A) 2CRT task and (B) 5CRT task

Data Processing

Two Choice Reaction Time Task (2CRT)

The RT in milliseconds (ms) was registered for each trial, and was measured as the time from target onset until a button response was identified. The button that was pressed for each trial was also recorded so that correct and incorrect responses could be identified. The data was filtered so that all RTs <150ms (deemed to be anticipatory rather than genuine responses) and all incorrect responses (i.e. where the participant pressed the wrong button) were excluded. Excessively slow trials (i.e. RTs > individual mean RT + 3SDs) were also removed and replaced with the mean RT over the remaining trials for each participant. These lower and upper bounds have been suggested by previous research (Hultsch, MacDonald, & Dixon, 2002; Hultsch et al., 2000) and removing such outliers and replacing missing values represents a conservative approach to computing IIV (Bunce et al., 2008). As a result of the filtering process, a total of 320 (3.1%) trials were excluded and 173 (1.7%) excessively slow trials were replaced.

Following the filtering of the data, the mean RT and RT IIV were calculated for each participant. Different measures have been proposed in the reporting of IIV (Hultsch et al., 2000). The raw standard deviation (SD) of responses has been used, but it has the disadvantage of being related to mean level of performance. A way of adjusting for this potentially confounding effect is to compute the coefficient of variation (CV), which involves dividing the raw SD by the mean RT (Hultsch & MacDonald, 2004) and it is this measure which has emerged as the standard measure of RT consistency recently due to its relatively easy calculation and high association with other measures of inconsistency (Bunce et al., 2013; Jackson et al., 2012; Lovden et al., 2007). Given this rationale, the CV was therefore calculated for each participant as a measure of IIV in this study. Finally, the number of incorrect responses for each participant was also calculated.

Five Choice Reaction Time Task (5CRT)

The RT in milliseconds (ms) was measured as the time from target onset until a touchscreen response was identified. The RT data was filtered in the same way as for the 2CRT task (see previous section). For this task, incorrect responses/task failures were defined as those trials where the participant touched the screen at a distance of >200 pixels from the target centre (these trials were outside the range of that deemed to be a “genuine” response). Trials were defined as “misses” when the participant touched the screen at a distance of >50 pixels but <200 pixels from the target centre (“misses” were classed as genuine attempts to touch the target and therefore included in the analysis). As a result of the filtering process, a total of 947 (9.1%) trials were excluded and 169 (1.6%) excessively slow trials were replaced.

In addition to mean RT, RT IIV and number of misses, spatial accuracy variables were also calculated for this task. Spatial accuracy was defined as the distance (measured in pixels) of the participant’s touch response from the centre of the target. Mean spatial accuracy and spatial accuracy IIV (calculated as the coefficient of variation: spatial accuracy SD/mean spatial accuracy) were calculated for each participant.

Statistical Analyses

All analyses were performed using SPSS Statistics v22 (IBM). Between-group differences in baseline characteristics (including age, years of education, NART IQ, GDS score and the neuropsychological test battery scores) were explored using the Kruskal-Wallis H test (since the data were not normally distributed). Subsequently, pairwise comparisons were performed using Dunn's (1964) procedure with a Bonferroni correction for multiple comparisons. Difference in gender proportion between the classification groups was analysed using the Chi-square test of homogeneity.

Mean RT and RT IIV for both tasks were also not normally distributed and an inverse transformation was applied to the data to achieve approximately normal distributions prior to subsequent analyses being run. One-way and mixed measures ANOVAs, with classification (cognitive status) as the between-group variable and task complexity (2CRT vs. 5CRT) as the within-group variable were used to explore for differences in mean RT and RT IIV between groups. Bonferroni corrected post hoc group-by-group comparisons were subsequently performed.

All accuracy variables (number of incorrect responses on the 2CRT task, number of misses, mean spatial accuracy and spatial accuracy IIV on the 5CRT task) were also not normally distributed. An inverse transformation was applied to mean spatial accuracy

and spatial accuracy IIV data to achieve approximately normal distributions prior to subsequent analyses being run. One way ANOVAs (with Bonferroni corrected post-hoc pair-wise comparisons) were applied to explore any between group-differences. Due to the nature of the data, the number of incorrect responses on the 2CRT task and the number of misses on the 5CRT could not be transformed and so between group differences were explored using the Kruskal-Wallis H test (with Bonferroni corrected post-hoc pairwise comparisons).

Cohen’s d effect sizes for any significant pairwise comparisons were calculated and were interpreted as follows: 0.2 < Cohen’s d < 0.5 = small, 0.5 < Cohen’s d < 0.8 = medium and 0.8 < Cohen’s d = large effect size (Cohen, 1988; Lakens, 2013).

Multinomial regressions were applied to assess whether the RT measures could predict classification. Age and NART IQ were included as covariates and the models were checked to ensure there was no multi-collinearity between the included variables. It is important to note that years of education was not included since it significantly correlated with both age and NART IQ.

Results

Baseline Characteristics

The baseline characteristics and neuropsychological battery test scores of the

participants by classification are shown in Table 5.1. Results from the Kruskal-Wallis H tests suggested that there were statistically significant differences between the groups in age (χ2(4) = 13.942, p=0.007), years of education (χ2(4) = 12.479, p=0.014) and NART IQ (χ2(4) = 10.755, p=0.029). However, post-hoc pair-wise comparisons revealed only a

significant difference in age between the controls and >MCI group (p=0.023), with the >MCI group being significantly older than the control group.

As the neuropsychological test scores formed the basis of the classifications, as expected, statistically significant differences were found between groups in GDS, CVLT short and long delay recall, Brixton Errors, Trail making tests A and B, VOSP- dot counting and -number location scores, CDT, GNT, BADLS (all at p<0.001) and VOSP-object decision score and PPT (at p<0.05) (see Table 5.1).

Table 5.1: Baseline characteristics and neuropsychological tests scores for the RT study cohort by classification

Control (n=164) aMCI (n=28) naMCI (n=11) >MCI (n=9) Low Mood (n = 12) p value

Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank Age (years) 75.0 (8) 104.67 77.5 (11) 125.79 81.0 (6) 144.32 80.0 (13)* 172.22 75.5 (12) 114.54 0.007 Gender (% female) 45 - 43 - 55 - 56 - 33 - 0.823 Education (years) 11 (3) 118.92 10 (2) 84.59 12 (2) 138.09 11 (1) 85.83 10 (2) 86.42 0.014 NART IQ† 116.0 (14) 119.74 112.5 (19) 99.27 115.0 (28) 100.27 105.0 (13) 78.56 106.5 (19) 72.42 0.029 GDS 1 (2) 99.94 2 (3)$ 119.34 2 (2)$ 131.05 2 (2) 156.06 7.5 (3)*** 218.50 <0.001

CVLT SD Free Recall, z score 0.5 (2) 134.98 -2.0 (1)*** 19.80 -0.5 (1) 82.18 -1.5 (1)*** 24.56 0.25 (3)# 115.25 <0.001 CVLT LD Free Recall, z score 0.5 (2) 135.10 -2.0 (1)*** 24.91 -1.0 (1)** 54.09 -2.0 (2)*** 35.83 0.50 (5)# 119.04 <0.001 Brixton Errors‡^ 18 (8) 97.70 22.5 (11) 131.88 32.5 (17)** 176.75 22.5 (18) 134.56 22.0 (12) 130.23 <0.001 Trails A†, percentile 60 (50) 127.38 30 (20)*** 71.61 10 (20)*** 45.41 10 (30)* 57.06 40 (45) 99.62 <0.001 Trails B§, percentile 60 (50) 123.93 25 (45)*** 67.92 10 (15)*** 29.90 15 (35)* 47.08 30 (35)* 69.04 <0.001

VOSP-Incomplete Letters score 20 (1) 115.71 20 (1) 117.88 19 (3) 76.27 19 (2) 102.22 19 (1) 97.00 0.204

VOSP-Object Decision score^ 19 (3) 120.35 18 (3) 99.48 17 (2) 77.36 17 (3) 68.00 18 (5) 101.12 0.018

VOSP-Dot Counting score 10 (0) 121.06 10 (1)** 87.48 10 (2)** 76.82 10 (2)* 79.22 10 (0) 111.58 <0.001

Control (n=164) aMCI (n=28) naMCI (n=11) >MCI (n=9) Low Mood (n = 12) p value Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank Median (IQR) Mean Rank

Clock Drawing Test score† 5 (0) 116.72 5 (0)% 109.55 5 (2) 95.59 4 (1.5)*** 54.89 5 (0)% 111.42 <0.001 Graded Naming Test score^ 24 (4) 126.85 19.5 (8)*** 71.95 17 (12) 76.82 19 (12)** 54.00 20 (8) 87.62 <0.001 Pyramids & Palm Trees Test score^ 51 (2) 122.69 50.5 (2) 88.73 49 (3) 71.45 51 (2) 86.28 50.5 (3) 86.04 0.002

BADLS score† 0 (0) 99.68 1 (3)**% 137.54 0 (2)% 119.23 14 (3)*** 219.00 0.5 (2)% 132.83 <0.001

KEY: BADLS = Bristol Activities of Daily Living Scale; CVLT LD = California Verbal Learning Test, long delay recall; CVLT SD = California Verbal Learning Test, short delay recall; IQR = interquartile range; VOSP = Visual Object and Space Perception

*p < 0.05, **p ≤ 0.01, ***p < 0.001 for difference from controls; #

p < 0.05 for difference from aMCI and >MCI; %p < 0.05 for difference from >MCI; $p < 0.05 for difference from Low Mood with Bonferroni correction for multiple comparisons

†

n = 163 control; ‡n = 161 control, 26 aMCI, 10 naMCI, 8 >MCI, 11 low mood; §n = 160 control, 26 aMCI, 10 naMCI, 6 >MCI